nb_kernel104_x86_64_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,085 行 · 第 1/5 页
S
2,085 行
addps %xmm1,%xmm0 addps %xmm2,%xmm0 addps %xmm4,%xmm3 addps %xmm5,%xmm3 addps %xmm7,%xmm6 addps %xmm8,%xmm6 ## start doing invsqrt for jM atoms rsqrtps %xmm0,%xmm1 rsqrtps %xmm3,%xmm4 rsqrtps %xmm6,%xmm7 movaps %xmm1,%xmm2 movaps %xmm4,%xmm5 movaps %xmm7,%xmm8 mulps %xmm1,%xmm1 ## lu*lu mulps %xmm4,%xmm4 ## lu*lu mulps %xmm7,%xmm7 ## lu*lu movaps nb104_three(%rsp),%xmm9 movaps %xmm9,%xmm10 movaps %xmm9,%xmm11 mulps %xmm0,%xmm1 ## rsq*lu*lu mulps %xmm3,%xmm4 ## rsq*lu*lu mulps %xmm6,%xmm7 ## rsq*lu*lu subps %xmm1,%xmm9 subps %xmm4,%xmm10 subps %xmm7,%xmm11 ## 3-rsq*lu*lu mulps %xmm2,%xmm9 mulps %xmm5,%xmm10 mulps %xmm8,%xmm11 ## lu*(3-rsq*lu*lu) movaps nb104_half(%rsp),%xmm0 mulps %xmm0,%xmm9 ## rinvH1M mulps %xmm0,%xmm10 ## rinvH2M mulps %xmm0,%xmm11 ## rinvMM ## M interactions movaps %xmm9,%xmm0 movaps %xmm10,%xmm1 movaps %xmm11,%xmm2 mulps %xmm9,%xmm9 mulps %xmm10,%xmm10 mulps %xmm11,%xmm11 mulps nb104_qqMH(%rsp),%xmm0 mulps nb104_qqMH(%rsp),%xmm1 mulps nb104_qqMM(%rsp),%xmm2 mulps %xmm0,%xmm9 mulps %xmm1,%xmm10 mulps %xmm2,%xmm11 addps nb104_vctot(%rsp),%xmm0 addps %xmm2,%xmm1 addps %xmm1,%xmm0 movaps %xmm0,nb104_vctot(%rsp) ## move j M forces to local temp variables movlps 36(%rdi,%rax,4),%xmm0 ## jxMa jyMa - - movlps 36(%rdi,%rcx,4),%xmm1 ## jxMc jyMc - - movhps 36(%rdi,%rbx,4),%xmm0 ## jxMa jyMa jxMb jyMb movhps 36(%rdi,%rdx,4),%xmm1 ## jxMc jyMc jxMd jyMd movss 44(%rdi,%rax,4),%xmm2 ## jzMa - - - movss 44(%rdi,%rcx,4),%xmm3 ## jzMc - - - movss 44(%rdi,%rbx,4),%xmm7 ## jzMb - - - movss 44(%rdi,%rdx,4),%xmm8 ## jzMd - - - movlhps %xmm7,%xmm2 ## jzMa - jzMb - movlhps %xmm8,%xmm3 ## jzMc - jzMd - shufps $136,%xmm3,%xmm2 ## 10001000 => jzMa jzMb jzMc jzMd ## xmm0: jxMa jyMa jxMb jyMb ## xmm1: jxMc jyMc jxMd jyMd ## xmm2: jzMa jzMb jzMc jzMd movaps %xmm9,%xmm7 movaps %xmm9,%xmm8 movaps %xmm11,%xmm13 movaps %xmm11,%xmm14 movaps %xmm11,%xmm15 movaps %xmm10,%xmm11 movaps %xmm10,%xmm12 mulps nb104_dxH1M(%rsp),%xmm7 mulps nb104_dyH1M(%rsp),%xmm8 mulps nb104_dzH1M(%rsp),%xmm9 mulps nb104_dxH2M(%rsp),%xmm10 mulps nb104_dyH2M(%rsp),%xmm11 mulps nb104_dzH2M(%rsp),%xmm12 mulps nb104_dxMM(%rsp),%xmm13 mulps nb104_dyMM(%rsp),%xmm14 mulps nb104_dzMM(%rsp),%xmm15 movaps %xmm7,%xmm3 movaps %xmm8,%xmm4 addps %xmm9,%xmm2 addps nb104_fixH1(%rsp),%xmm7 addps nb104_fiyH1(%rsp),%xmm8 addps nb104_fizH1(%rsp),%xmm9 addps %xmm10,%xmm3 addps %xmm11,%xmm4 addps %xmm12,%xmm2 addps nb104_fixH2(%rsp),%xmm10 addps nb104_fiyH2(%rsp),%xmm11 addps nb104_fizH2(%rsp),%xmm12 addps %xmm13,%xmm3 addps %xmm14,%xmm4 addps %xmm15,%xmm2 addps nb104_fixM(%rsp),%xmm13 addps nb104_fiyM(%rsp),%xmm14 addps nb104_fizM(%rsp),%xmm15 movaps %xmm7,nb104_fixH1(%rsp) movaps %xmm8,nb104_fiyH1(%rsp) movaps %xmm9,nb104_fizH1(%rsp) movaps %xmm10,nb104_fixH2(%rsp) movaps %xmm11,nb104_fiyH2(%rsp) movaps %xmm12,nb104_fizH2(%rsp) movaps %xmm13,nb104_fixM(%rsp) movaps %xmm14,nb104_fiyM(%rsp) movaps %xmm15,nb104_fizM(%rsp) ## xmm3 = fMx , xmm4 = fMy movaps %xmm3,%xmm5 unpcklps %xmm4,%xmm3 unpckhps %xmm4,%xmm5 addps %xmm3,%xmm0 addps %xmm5,%xmm1 movhlps %xmm2,%xmm3 ## fMzc fMzd movlps %xmm0,36(%rdi,%rax,4) movhps %xmm0,36(%rdi,%rbx,4) movlps %xmm1,36(%rdi,%rcx,4) movhps %xmm1,36(%rdi,%rdx,4) movss %xmm2,44(%rdi,%rax,4) movss %xmm3,44(%rdi,%rcx,4) shufps $1,%xmm2,%xmm2 shufps $1,%xmm3,%xmm3 movss %xmm2,44(%rdi,%rbx,4) movss %xmm3,44(%rdi,%rdx,4) ## should we do one more iteration? subl $4,nb104_innerk(%rsp) jl _nb_kernel104_x86_64_sse.nb104_single_check jmp _nb_kernel104_x86_64_sse.nb104_unroll_loop_nb_kernel104_x86_64_sse.nb104_single_check: addl $4,nb104_innerk(%rsp) jnz _nb_kernel104_x86_64_sse.nb104_single_loop jmp _nb_kernel104_x86_64_sse.nb104_updateouterdata_nb_kernel104_x86_64_sse.nb104_single_loop: movq nb104_innerjjnr(%rsp),%rdx ## pointer to jjnr[k] movl (%rdx),%eax addq $4,nb104_innerjjnr(%rsp) movq nb104_pos(%rbp),%rsi lea (%rax,%rax,2),%rax ## fetch j coordinates xorps %xmm0,%xmm0 xorps %xmm1,%xmm1 xorps %xmm2,%xmm2 movss 36(%rsi,%rax,4),%xmm0 ## jxM - - - movss 40(%rsi,%rax,4),%xmm1 ## jyM - - - movss 44(%rsi,%rax,4),%xmm2 ## jzM - - - movlps 12(%rsi,%rax,4),%xmm6 ## xmm6 = jxH1 jyH1 - - movss 20(%rsi,%rax,4),%xmm7 ## xmm7 = jzH1 - - - movhps 24(%rsi,%rax,4),%xmm6 ## xmm6 = jxH1 jyH1 jxH2 jyH2 movss 32(%rsi,%rax,4),%xmm5 ## xmm5 = jzH2 - - - ## have all coords, time for some shuffling. shufps $216,%xmm6,%xmm6 ## 11011000 ;# xmm6 = jxH1 jxH2 jyH1 jyH2 unpcklps %xmm5,%xmm7 ## xmm7 = jzH1 jzH2 - - movlhps %xmm6,%xmm0 ## xmm0 = jxM 0 jxH1 jxH2 shufps $228,%xmm6,%xmm1 ## 11100100 ;# xmm1 = jyM 0 jyH1 jyH2 shufps $68,%xmm7,%xmm2 ## 01000100 ;# xmm2 = jzM 0 jzH1 jzH2 ## store all j coordinates in jM movaps %xmm0,nb104_jxM(%rsp) movaps %xmm1,nb104_jyM(%rsp) movaps %xmm2,nb104_jzM(%rsp) subps nb104_ixM(%rsp),%xmm0 subps nb104_iyM(%rsp),%xmm1 subps nb104_izM(%rsp),%xmm2 movaps %xmm0,nb104_dxMM(%rsp) movaps %xmm1,nb104_dyMM(%rsp) movaps %xmm2,nb104_dzMM(%rsp) mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 addps %xmm1,%xmm0 addps %xmm2,%xmm0 ## have rsq in xmm0 ## do invsqrt rsqrtps %xmm0,%xmm1 movaps %xmm1,%xmm2 mulps %xmm1,%xmm1 movaps nb104_three(%rsp),%xmm3 mulps %xmm0,%xmm1 subps %xmm1,%xmm3 mulps %xmm2,%xmm3 mulps nb104_half(%rsp),%xmm3 ## rinv iM- j water xorps %xmm1,%xmm1 movaps %xmm3,%xmm0 xorps %xmm4,%xmm4 mulps %xmm0,%xmm0 ## xmm0=rinvsq ## fetch charges to xmm4 movss nb104_qqMM(%rsp),%xmm4 movhps nb104_qqMH(%rsp),%xmm4 mulps %xmm4,%xmm3 ## xmm3=vcoul mulps %xmm3,%xmm0 ## total fscal addps nb104_vctot(%rsp),%xmm3 movaps %xmm3,nb104_vctot(%rsp) movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb104_dxMM(%rsp),%xmm0 mulps nb104_dyMM(%rsp),%xmm1 mulps nb104_dzMM(%rsp),%xmm2 ## initial update for j forces xorps %xmm3,%xmm3 xorps %xmm4,%xmm4 xorps %xmm5,%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb104_fjxM(%rsp) movaps %xmm4,nb104_fjyM(%rsp) movaps %xmm5,nb104_fjzM(%rsp) addps nb104_fixM(%rsp),%xmm0 addps nb104_fiyM(%rsp),%xmm1 addps nb104_fizM(%rsp),%xmm2 movaps %xmm0,nb104_fixM(%rsp) movaps %xmm1,nb104_fiyM(%rsp) movaps %xmm2,nb104_fizM(%rsp) ## done with i M Now do i H1 & H2 simultaneously first get i particle coords: movaps nb104_jxM(%rsp),%xmm0 movaps nb104_jyM(%rsp),%xmm1 movaps nb104_jzM(%rsp),%xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 subps nb104_ixH1(%rsp),%xmm0 subps nb104_iyH1(%rsp),%xmm1 subps nb104_izH1(%rsp),%xmm2 subps nb104_ixH2(%rsp),%xmm3 subps nb104_iyH2(%rsp),%xmm4 subps nb104_izH2(%rsp),%xmm5 movaps %xmm0,nb104_dxH1M(%rsp) movaps %xmm1,nb104_dyH1M(%rsp) movaps %xmm2,nb104_dzH1M(%rsp) movaps %xmm3,nb104_dxH2M(%rsp) movaps %xmm4,nb104_dyH2M(%rsp) movaps %xmm5,nb104_dzH2M(%rsp) mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 addps %xmm1,%xmm0 addps %xmm3,%xmm4 addps %xmm2,%xmm0 ## have rsqH1 in xmm0 addps %xmm5,%xmm4 ## have rsqH2 in xmm4 ## do invsqrt rsqrtps %xmm0,%xmm1 rsqrtps %xmm4,%xmm5 movaps %xmm1,%xmm2 movaps %xmm5,%xmm6 mulps %xmm1,%xmm1 mulps %xmm5,%xmm5 movaps nb104_three(%rsp),%xmm3 movaps %xmm3,%xmm7 mulps %xmm0,%xmm1 mulps %xmm4,%xmm5 subps %xmm1,%xmm3 subps %xmm5,%xmm7 mulps %xmm2,%xmm3 mulps %xmm6,%xmm7 mulps nb104_half(%rsp),%xmm3 ## rinv H1 - j water mulps nb104_half(%rsp),%xmm7 ## rinv H2 - j water ## assemble charges in xmm6 xorps %xmm6,%xmm6 movss nb104_qqMH(%rsp),%xmm6 movhps nb104_qqHH(%rsp),%xmm6 ## do coulomb interaction movaps %xmm3,%xmm0 movaps %xmm7,%xmm4 mulps %xmm0,%xmm0 ## rinvsq mulps %xmm4,%xmm4 ## rinvsq mulps %xmm6,%xmm3 ## vcoul mulps %xmm6,%xmm7 ## vcoul movaps %xmm3,%xmm2 addps %xmm7,%xmm2 ## total vcoul mulps %xmm3,%xmm0 ## fscal addps nb104_vctot(%rsp),%xmm2 mulps %xmm4,%xmm7 ## fscal movaps %xmm2,nb104_vctot(%rsp) movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb104_dxH1M(%rsp),%xmm0 mulps nb104_dyH1M(%rsp),%xmm1 mulps nb104_dzH1M(%rsp),%xmm2 ## update forces H1 - j water movaps nb104_fjxM(%rsp),%xmm3 movaps nb104_fjyM(%rsp),%xmm4 movaps nb104_fjzM(%rsp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb104_fjxM(%rsp) movaps %xmm4,nb104_fjyM(%rsp) movaps %xmm5,nb104_fjzM(%rsp) addps nb104_fixH1(%rsp),%xmm0 addps nb104_fiyH1(%rsp),%xmm1 addps nb104_fizH1(%rsp),%xmm2 movaps %xmm0,nb104_fixH1(%rsp) movaps %xmm1,nb104_fiyH1(%rsp) movaps %xmm2,nb104_fizH1(%rsp) ## do forces H2 - j water movaps %xmm7,%xmm0 movaps %xmm7,%xmm1 movaps %xmm7,%xmm2 mulps nb104_dxH2M(%rsp),%xmm0 mulps nb104_dyH2M(%rsp),%xmm1 mulps nb104_dzH2M(%rsp),%xmm2 movaps nb104_fjxM(%rsp),%xmm3 movaps nb104_fjyM(%rsp),%xmm4 movaps nb104_fjzM(%rsp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movq nb104_faction(%rbp),%rsi movaps %xmm3,nb104_fjxM(%rsp) movaps %xmm4,nb104_fjyM(%rsp) movaps %xmm5,nb104_fjzM(%rsp) addps nb104_fixH2(%rsp),%xmm0 addps nb104_fiyH2(%rsp),%xmm1 addps nb104_fizH2(%rsp),%xmm2 movaps %xmm0,nb104_fixH2(%rsp) movaps %xmm1,nb104_fiyH2(%rsp) movaps %xmm2,nb104_fizH2(%rsp) ## update j water forces from local variables movlps 36(%rsi,%rax,4),%xmm0 movlps 12(%rsi,%rax,4),%xmm1 movhps 24(%rsi,%rax,4),%xmm1 movaps nb104_fjxM(%rsp),%xmm3 movaps nb104_fjyM(%rsp),%xmm4 movaps nb104_fjzM(%rsp),%xmm5 movaps %xmm5,%xmm6 movaps %xmm5,%xmm7 shufps $2,%xmm6,%xmm6 ## 00000010 shufps $3,%xmm7,%xmm7 ## 00000011 addss 44(%rsi,%rax,4),%xmm5 addss 20(%rsi,%rax,4),%xmm6 addss 32(%rsi,%rax,4),%xmm7 movss %xmm5,44(%rsi,%rax,4) movss %xmm6,20(%rsi,%rax,4) movss %xmm7,32(%rsi,%rax,4) movaps %xmm3,%xmm5 unpcklps %xmm4,%xmm3 unpckhps %xmm4,%xmm5 addps %xmm3,%xmm0 addps %xmm5,%xmm1 movlps %xmm0,36(%rsi,%rax,4) movlps %xmm1,12(%rsi,%rax,4) movhps %xmm1,24(%rsi,%rax,4) decl nb104_innerk(%rsp) jz _nb_kernel104_x86_64_sse.nb104_updateouterdata jmp _nb_kernel104_x86_64_sse.nb104_single_loop_nb_kernel104_x86_64_sse.nb104_updateouterdata: movl nb104_ii3(%rsp),%ecx movq nb104_faction(%rbp),%rdi movq nb104_fshift(%rbp),%rsi movl nb104_is3(%rsp),%edx ## accumulate H1i forces in xmm0, xmm1, xmm2 movaps nb104_fixH1(%rsp),%xmm0 movaps nb104_fiyH1(%rsp),%xmm1 movaps nb104_fizH1(%rsp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addps %xmm3,%xmm0 addps %xmm4,%xmm1 addps %xmm5,%xmm2 ## sum is in 1/2 in xmm0-xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 shufps $1,%xmm3,%xmm3⌨️ 快捷键说明
复制代码Ctrl + C
搜索代码Ctrl + F
全屏模式F11
增大字号Ctrl + =
减小字号Ctrl + -
显示快捷键?